The Curse of Class Imbalance and Conflicting Metrics with Machine Learning for Side-channel Evaluations

Stjepan Picek,Alan Jovic,Shivam Bhasin,Annelie Heuser,Francesco Regazzoni

doi:10.46586/tches.v2019.i1.209-237

Abstract

We concentrate on machine learning techniques used for profiled sidechannel analysis in the presence of imbalanced data. Such scenarios are realistic and often occurring, for instance in the Hamming weight or Hamming distance leakage models. In order to deal with the imbalanced data, we use various balancing techniques and we show that most of them help in mounting successful attacks when the data is highly imbalanced. Especially, the results with the SMOTE technique are encouraging, since we observe some scenarios where it reduces the number of necessary measurements more than 8 times. Next, we provide extensive results on comparison of machine learning and side-channel metrics, where we show that machine learning metrics (and especially accuracy as the most often used one) can be extremely deceptive. This finding opens a need to revisit the previous works and their results in order to properly assess the performance of machine learning in side-channel analysis.

Highlights

Side-channel Attacks (SCA) is a serious threat, which exploits weaknesses in the physical implementation of cryptographic algorithms [MOP06]
Our results clearly demonstrate that, if the classification problem is sufficiently hard and there is an imbalance within the dataset, data sampling techniques may increase Success rate (SR) and Guessing entropy (GE) significantly
On a more general level, our experiments indicate that none of the machine learning (ML) metrics we tested can be used as a reliable indicator of SCA performance when dealing with imbalanced data

Summary

Introduction

Side-channel Attacks (SCA) is a serious threat, which exploits weaknesses in the physical implementation of cryptographic algorithms [MOP06]. A data transition from 0 → 1 or 1 → 0 in a CMOS cell causes current flow leading to power consumption. This can be distinguished from the case when no transition occurs (0 → 0 or 1 → 1). The adversary obtains side-channel measurements from a clone device with known inputs, including the secret key. From this data set, known as the profiling set, the adversary completely characterizes the relevant leakages. Characterized leakages are typically obtained for the secret key dependent intermediate values, that are processed on the device and result in physical leakages. These models can be used in the attacking phase on the target device to predict which intermediate values are processed, revealing information about the secret key

Results

Discussion

Conclusion